source: simon willison: quoting james shore
level: technical
james shore argues that ai coding agents need to lower maintenance costs, not just speed up writing code. if a tool makes you twice as productive, your maintenance costs should be halved. otherwise, the temporary speed boost leads to permanent extra work. the math is simple: doubling output without reducing maintenance costs doubles the total maintenance burden. if output triples, maintenance costs must drop to one third to break even.
the core problem is that faster code generation can hide growing technical debt. when developers produce more code quickly, they also create more code to maintain. without a matching drop in maintenance effort, the long-term cost multiplies. shore points out that doubling output while holding maintenance steady still doubles maintenance costs. the relationship is inverse: the productivity multiplier must be matched by a maintenance cost divider.
this warning applies to any team adopting ai coding assistants. the initial speed feels like a win, but the real test is whether the generated code is easier to maintain. if not, teams trade short-term velocity for long-term drag. shore's message is clear: measure maintenance costs alongside productivity gains. only tools that reduce the cost of keeping code healthy over time are worth using.
why it matters: for ai and data science teams, this highlights the risk of adopting coding agents without measuring their impact on long-term code health and maintenance workload.